Band Grouping SuperPCA for Feature Extraction and Extended Morphological Profile Production From Hyperspectral Images
Behnam Asghari Beirami, Mehdi Mokhtarzade
Abstract
A variety of dimensionality reduction methods have been proposed in the literature to address the issue of information redundancy in hyperspectral images. A recent one, named superpixel-based principal components analysis (SuperPCA), integrates the superpixel segmentation algorithm and principal components analysis (PCA) to exploit contextual information in the process of feature extraction. The present letter aims at improving SuperPCA method via a band grouping technique. It also provides appropriate base images for the production of extended morphological profiles (EMPs). This proposed method was applied to two real case studies and the results proved an overall accuracy improvement of 8% over the SuperPCA. Furthermore, the EMPs resulted from our proposed method could reach to better spatial-spectral classification results in comparison to those methods that apply conventional base images for EMPs production.